From Symptoms to Services: An LLM Chatbot for Effective Departmental Referral

Abstract

This study explores integrating large language models (LLM) into the medical domain, focusing on developing and using the LLM tool Chat-SymChecker. Although LLM technology, such as ChatGPT and Copilot, is evolving rapidly, their use in medical consultations is still limited due to their complexity. To address this issue, we propose using Chat-Symptom Checker to supplement primary care consultation and specialist referral. Chat-Symptom Checker is based on the LLaMA model and is trained on extensive medical Question-answer datasets and patient-specific data from electronic health records, allowing it to provide rapid initial assessment and efficiently direct patients to the proper medical department. This article describes Chat-Symptom Checker's development process, functionality, and potential impact in increasing hospital efficiency, accelerating diagnostic procedures, and enhancing patient care. Chat-Symptom Checker shows the capability of processing complex natural language input, allowing users to describe symptoms and receive clear, individualized feedback. By integrating comprehensive patient data, such as past medical history and family history, the system will guide users to the proper medical department and provide initial recommendations and potential diagnoses, which can significantly decrease wait times and labor costs while simultaneously improving service efficiency. However, there are several challenges with our model. Issues such as redundant or nonsensical queries still need to be refined. In addition, an evaluation of the text quality of LLMs reveals that data volume is not necessarily correlated with enhanced performance. Studies reveal that smaller datasets with better text quality can perform better than more enormous datasets that lack context coherence. This shows the significance of data quality and contextual relevance in LLM medical model training. Despite some remaining limitations, Chat-Symptom Checker can serve as a beneficial healthcare support tool.